Development and temporal validation of a machine learning-based prediction model for long-term depressive symptoms in older adults with cardiovascular disease or hypertension

Figures

• 

  Figure 1.. Flowchart of the study sample selection. CHARLS = China Health and Retirement Longitudinal Study, CVD = cardiovascular disease.

• 

  Figure 2.. Feature selection using the Boruta algorithm. Green box plots represent confirmed predictors, yellow box plots represent tentative predictors, red box plots represent rejected predictors, and blue box plots represent shadow features. HDL = high-density lipoprotein cholesterol, LDL = low-density lipoprotein cholesterol, TyG = triglyceride-glucose.

• 

  Figure 3.. Predictive performance of 7 machine learning models: (A) receiver operating characteristic curves; (B) decision curves; (C) calibration curves; and (D) precision-recall curves. In the software-generated panel titles, “Validation” refers to the development cohort. AP = average precision, AUC = area under the receiver operating characteristic curve, GBDT = gradient boosting decision tree, KNN = k-nearest neighbors, PR = precision-recall, ROC = receiver operating characteristic, SVM = support vector machine, XGBoost = extreme gradient boosting.

• 

  Figure 4.. Internal and temporal validation results of the logistic regression model: (A) 10-fold cross-validation in the development cohort; and (B) temporal validation in the 2015 to 2020 cohort. AUC = area under the receiver operating characteristic curve, CI = confidence interval, ROC = receiver operating characteristic.

• 

  Figure 5.. K–S statistic plot for the temporal validation cohort. K–S = Kolmogorov–Smirnov.

• 

  Figure 6.. Confusion matrix for the temporal validation cohort.

• 

  Figure 7.. SHAP (Shapley additive explanations) analysis of the model. (A) SHAP summary plot visualizing the distribution of SHAP values for each feature. Each dot represents an individual data point, with the x-axis showing the SHAP value (feature’s impact on the model output) and the color representing the feature value (red for higher values, blue for lower values). (B) Bar plot showing the mean absolute SHAP values for all 7 predictors ranked by their contribution to predicting depressive symptoms in older adults with CVD or hypertension. CVD = cardiovascular disease, TyG = triglyceride-glucose.

• 

  Figure 8.. SHAP dependence plots: (A) total household income; (B) cognitive score; (C) average daily sleep duration; (D) mean right-hand grip strength; and (E) TyG index. SHAP = Shapley additive explanations, TyG = triglyceride-glucose.

• 

  Figure 9.. The dynamic nomogram based on the logistic regression model. The nomogram used sex, educational attainment, TyG index, cognitive score, total household income, average daily sleep duration, and mean right-hand grip strength to give a person-specific prediction of the probability of developing depressive symptoms. The red point represents the predictor value of an example person, and the blue lines or boxes represent the distribution of the variable being studied in the sample population. Predictor contributions were summed to obtain the total score; in the given scenario, a score of 1.26 was associated with a predicted probability of 27.3%. TyG = triglyceride-glucose.